"Percolation on transitive graphs", Ottawa
Transcript of "Percolation on transitive graphs", Ottawa
Percolation on transitive graphs
Mark Sapir
A short survey plus Iva Kozakova’s work
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
For every p ∈ (0, 1), the Bernoulli bond percolation on G
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
For every p ∈ (0, 1), the Bernoulli bond percolation on Gis theproduct probability measure Pp on the space of subsets of E .
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
For every p ∈ (0, 1), the Bernoulli bond percolation on Gis theproduct probability measure Pp on the space of subsets of E .
The product measure is defined as follows. The σ-algebra isgenerated by the cylindrical subset Es , where Es is the set ofsubsets of E containing a given edge s.
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
For every p ∈ (0, 1), the Bernoulli bond percolation on Gis theproduct probability measure Pp on the space of subsets of E .
The product measure is defined as follows. The σ-algebra isgenerated by the cylindrical subset Es , where Es is the set ofsubsets of E containing a given edge s.
The measure is given by Pp(Es) = p.
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
For every p ∈ (0, 1), the Bernoulli bond percolation on Gis theproduct probability measure Pp on the space of subsets of E .
The product measure is defined as follows. The σ-algebra isgenerated by the cylindrical subset Es , where Es is the set ofsubsets of E containing a given edge s.
The measure is given by Pp(Es) = p.
For any realization ω ∈ Ω, open edges form a random subgraph ofG.
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
For every p ∈ (0, 1), the Bernoulli bond percolation on Gis theproduct probability measure Pp on the space of subsets of E .
The product measure is defined as follows. The σ-algebra isgenerated by the cylindrical subset Es , where Es is the set ofsubsets of E containing a given edge s.
The measure is given by Pp(Es) = p.
For any realization ω ∈ Ω, open edges form a random subgraph ofG. Connected components of that subgraph are called clusters.
The definition
Consider a transitive locally finite graph G = (V , E ) with origin o.
For every p ∈ (0, 1), the Bernoulli bond percolation on Gis theproduct probability measure Pp on the space of subsets of E .
The product measure is defined as follows. The σ-algebra isgenerated by the cylindrical subset Es , where Es is the set ofsubsets of E containing a given edge s.
The measure is given by Pp(Es) = p.
For any realization ω ∈ Ω, open edges form a random subgraph ofG. Connected components of that subgraph are called clusters.ThePercolation function θ(p) is defined to be the probability that theorigin is contained in an infinite cluster.
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law:
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Example of such an event:
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Example of such an event:
existence of an infinite cluster.
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Example of such an event:
existence of an infinite cluster.
existence of k infinite clusters (for any given k).
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Example of such an event:
existence of an infinite cluster.
existence of k infinite clusters (for any given k).
Therefore if the graph is transitive, then
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Example of such an event:
existence of an infinite cluster.
existence of k infinite clusters (for any given k).
Therefore if the graph is transitive, then either with probability 1there are no infinite clusters (that is p < pc), or
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Example of such an event:
existence of an infinite cluster.
existence of k infinite clusters (for any given k).
Therefore if the graph is transitive, then either with probability 1there are no infinite clusters (that is p < pc), or with probability 1,there exists a unique infinite cluster or
Why transitive
For transitive graphs, we have the Kolmogorov 0-1 law: every tailinvariant event happens with probability 0 or 1.
Example of such an event:
existence of an infinite cluster.
existence of k infinite clusters (for any given k).
Therefore if the graph is transitive, then either with probability 1there are no infinite clusters (that is p < pc), or with probability 1,there exists a unique infinite cluster or with probability 1 there areinfinitely many of them.
The critical value pc
The critical value pc of percolation
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The standard square lattice Z2; pc = 12 (Kesten);
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The standard square lattice Z2; pc = 12 (Kesten);
The triangular lattice in R2; pc = 2 sin(π/18);
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The standard square lattice Z2; pc = 12 (Kesten);
The triangular lattice in R2; pc = 2 sin(π/18);
The hexagonal lattice in R2; pc = 1 − 2 sin(π/18);
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The standard square lattice Z2; pc = 12 (Kesten);
The triangular lattice in R2; pc = 2 sin(π/18);
The hexagonal lattice in R2; pc = 1 − 2 sin(π/18);
Some other lattices in R2, using duality.
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The standard square lattice Z2; pc = 12 (Kesten);
The triangular lattice in R2; pc = 2 sin(π/18);
The hexagonal lattice in R2; pc = 1 − 2 sin(π/18);
Some other lattices in R2, using duality.
Trees; pc = 1/branching number (Lyons).
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The standard square lattice Z2; pc = 12 (Kesten);
The triangular lattice in R2; pc = 2 sin(π/18);
The hexagonal lattice in R2; pc = 1 − 2 sin(π/18);
Some other lattices in R2, using duality.
Trees; pc = 1/branching number (Lyons).
Cayley graphs of virtually cyclic groups; pc = 1
The critical value pc
The critical value pc of percolation is such that for 0 ≤ p < pc ,θ(p) = 0, and if pc < p ≤ 1, then θ(p) > 0.
Explicit values of pc are known only in the following cases:
The standard square lattice Z2; pc = 12 (Kesten);
The triangular lattice in R2; pc = 2 sin(π/18);
The hexagonal lattice in R2; pc = 1 − 2 sin(π/18);
Some other lattices in R2, using duality.
Trees; pc = 1/branching number (Lyons).
Cayley graphs of virtually cyclic groups; pc = 1
Free products of transitive graphs; pc can be expressed interms of the expected cluster sizes at the origin of the freefactors. (Kozakova)
Other critical characteristics of percolation
The uniqueness phase pu - the infimum of all p such that Pp-a.s.the infinite cluster is unique. By Haggstrom and Peres, the infinitecluster is unique a.s. for all p > pu.
It is known that pc ≤ pu for every group.
For amenable groups, pc = pu.
For the Cayley graph of F2, pc = 13 , pu = 1. For every transitive
graph with infinitely many ends, pu = 1.
Other critical characteristics of percolation
The uniqueness phase pu - the infimum of all p such that Pp-a.s.the infinite cluster is unique. By Haggstrom and Peres, the infinitecluster is unique a.s. for all p > pu.
It is known that pc ≤ pu for every group.
For amenable groups, pc = pu.
For the Cayley graph of F2, pc = 13 , pu = 1. For every transitive
graph with infinitely many ends, pu = 1.
Pak and Smirnova-Nagnibeda:
Other critical characteristics of percolation
The uniqueness phase pu - the infimum of all p such that Pp-a.s.the infinite cluster is unique. By Haggstrom and Peres, the infinitecluster is unique a.s. for all p > pu.
It is known that pc ≤ pu for every group.
For amenable groups, pc = pu.
For the Cayley graph of F2, pc = 13 , pu = 1. For every transitive
graph with infinitely many ends, pu = 1.
Pak and Smirnova-Nagnibeda:For every non-amenable group, thereexists a generating set, possibly with repetitions, for which pc < pu.
Cost of non-amenable groups
Gaboriau: The first ℓ2-Betti number of a group does not exceed12(pu − pc).
Cost of non-amenable groups
Gaboriau: The first ℓ2-Betti number of a group does not exceed12(pu − pc).
pu is related to the cost of a group and to the “measurable”version of the von Neumann conjecture by Gaboreau and Lyons .
Cost of non-amenable groups
Gaboriau: The first ℓ2-Betti number of a group does not exceed12(pu − pc).
pu is related to the cost of a group and to the “measurable”version of the von Neumann conjecture by Gaboreau and Lyons .
Theorem (Gaboreau, Lyons) For any finitely generatednon-amenable group G , there is n ∈ N and a non-empty interval(p1, p2) of parameters p for which there is an ergodic essentiallyfree action of F2 on Πn
1(0, 1G , µp) such that almost everyG -orbit of the diagonal Bernoulli shift decomposes into F2-orbits.
Schonmann’s critical value
Schonmann’s critical value of p:
pexp = supp : ∃C ,γ>0∀x ,y∈V Pp(x ↔ y) ≤ Ce−γdist(x ,y)
Schonmann’s critical value
Schonmann’s critical value of p:
pexp = supp : ∃C ,γ>0∀x ,y∈V Pp(x ↔ y) ≤ Ce−γdist(x ,y)
pc ≤ pexp ≤ pu.
Critical exponents
The value of the critical probabilities is not a quasi-isometryinvariant.
Critical exponents
The value of the critical probabilities is not a quasi-isometryinvariant.
The behavior of percolation at pc is conjectured to be more stable.
Critical exponents
The value of the critical probabilities is not a quasi-isometryinvariant.
The behavior of percolation at pc is conjectured to be more stable.
Assume that
θ(p) ≈ (p − pc)β as p ց pc ,
χ(p) ≈ (pc − p)γ as p ր pc .
Then we say that β and γ are critical exponents.
Critical exponents
The value of the critical probabilities is not a quasi-isometryinvariant.
The behavior of percolation at pc is conjectured to be more stable.
Assume that
θ(p) ≈ (p − pc)β as p ց pc ,
χ(p) ≈ (pc − p)γ as p ր pc .
Then we say that β and γ are critical exponents.For graphs with infinitely many ends, β = 1, γ = −1. These arecalled the mean field values.
Critical exponents
The value of the critical probabilities is not a quasi-isometryinvariant.
The behavior of percolation at pc is conjectured to be more stable.
Assume that
θ(p) ≈ (p − pc)β as p ց pc ,
χ(p) ≈ (pc − p)γ as p ր pc .
Then we say that β and γ are critical exponents.For graphs with infinitely many ends, β = 1, γ = −1. These arecalled the mean field values.
For the triangular lattice in R2 we have β = 5/36, γ = −43/18 asproved by Smirnov and Werner.
The main problems about percolation
Problem (Benjamini-Schramm)
Is it true that pc 6= pu for every Cayley graph of a non-amenable
group?
The main problems about percolation
Problem (Benjamini-Schramm)
Is it true that pc 6= pu for every Cayley graph of a non-amenable
group?
Schonmann: pc < pu for highly nonamenable groups (i.e. suchthat the Cheeger constant is bigger than (
√2d2 − 1 − 1)/2, where
d is the vertex degree).
The main problems about percolation
Problem (Benjamini-Schramm)
Is it true that pc 6= pu for every Cayley graph of a non-amenable
group?
Schonmann: pc < pu for highly nonamenable groups (i.e. suchthat the Cheeger constant is bigger than (
√2d2 − 1 − 1)/2, where
d is the vertex degree).
ProblemFind pc for the cubic lattice in R3?
Main problems continued
Problem (Benjamini, Schramm)
Assume that G has growth function faster than linear, is pc < 1?
Main problems continued
Problem (Benjamini, Schramm)
Assume that G has growth function faster than linear, is pc < 1?
Only amenable case is of interest, since pc ≤ 1iE+1 .
Main problems continued
Problem (Benjamini, Schramm)
Assume that G has growth function faster than linear, is pc < 1?
Only amenable case is of interest, since pc ≤ 1iE+1 .
Lyons: for groups with polynomial or exponential growth pc < 1.The same is true for all finitely presented groups(Babson+Benjamini).
Main problems continued
Problem (Benjamini, Schramm)
Assume that G has growth function faster than linear, is pc < 1?
Only amenable case is of interest, since pc ≤ 1iE+1 .
Lyons: for groups with polynomial or exponential growth pc < 1.The same is true for all finitely presented groups(Babson+Benjamini).
Problem (Smirnova-Nagnibeda)
Find pc of known groups with “standard” generating sets.
Distortion of clusters
ProblemLet p > pc . What is the distortion of an infinite cluster?
Distortion of clusters
ProblemLet p > pc . What is the distortion of an infinite cluster?
On trees the distortion is linear.
Distortion of clusters
ProblemLet p > pc . What is the distortion of an infinite cluster?
On trees the distortion is linear.
ProblemWhat is the possible expected distortion of an open cluster in a
Cayley graph of a group? Same question for hyperbolic groups is
also open.
Critical exponents
Hara and Slade: Zd has meanfield critical exponents for d ≥ 19.
Critical exponents
Hara and Slade: Zd has meanfield critical exponents for d ≥ 19.
ProblemIs it true that the critical exponents of all Cayley graphs of groups
with isometric asymptotic cones are equal? In particular, is it true
that every Cayley graph of a non-elementary hyperbolic group has
mean-field valued critical exponents?
Critical exponents
Hara and Slade: Zd has meanfield critical exponents for d ≥ 19.
ProblemIs it true that the critical exponents of all Cayley graphs of groups
with isometric asymptotic cones are equal? In particular, is it true
that every Cayley graph of a non-elementary hyperbolic group has
mean-field valued critical exponents?
Schonmann: the critical exponents take their mean-field values forall non-amenable planar graphs with one end, and for unimodulargraphs with infinitely many ends (in particular, for all Cayleygraphs of groups with infinitely many ends).
Rescaling
Instead of considering the limit as p → pc , consider tessellations ofZd by cubes with bigger and bigger sizes.
Rescaling
Instead of considering the limit as p → pc , consider tessellations ofZd by cubes with bigger and bigger sizes.An edge is open if one can cross the cube in the direction of theedge using open edges of Zd .
Rescaling
Instead of considering the limit as p → pc , consider tessellations ofZd by cubes with bigger and bigger sizes.An edge is open if one can cross the cube in the direction of theedge using open edges of Zd .
Problem (Benjamini)
Is it true that every scale invariant finitely generated group is
virtually nilpotent?
Rescaling
Instead of considering the limit as p → pc , consider tessellations ofZd by cubes with bigger and bigger sizes.An edge is open if one can cross the cube in the direction of theedge using open edges of Zd .
Problem (Benjamini)
Is it true that every scale invariant finitely generated group is
virtually nilpotent?
Pete+Nekrashevich+S: Z2wrZ .
Kozakova’s results I
Let G = G1 ∗ G2. Then the Cayley graph is the free product ofgraphs of G1 and G2.
Kozakova’s results I
Let G = G1 ∗ G2. Then the Cayley graph is the free product ofgraphs of G1 and G2.Consider the branching process with two types of children.
Kozakova’s results I
Let G = G1 ∗ G2. Then the Cayley graph is the free product ofgraphs of G1 and G2.Consider the branching process with two types of children. The
Galton-Watson matrix is
(
0 χ1(p) − 1χ2(p) − 1 0
)
. The
population dies out iff the spectral radius is 1. This gives the firstpart of
Kozakova’s results I
Let G = G1 ∗ G2. Then the Cayley graph is the free product ofgraphs of G1 and G2.Consider the branching process with two types of children. The
Galton-Watson matrix is
(
0 χ1(p) − 1χ2(p) − 1 0
)
. The
population dies out iff the spectral radius is 1. This gives the firstpart of
Theorem (Kozakova)
Let G = G1 ∗ G2. Then
Kozakova’s results I
Let G = G1 ∗ G2. Then the Cayley graph is the free product ofgraphs of G1 and G2.Consider the branching process with two types of children. The
Galton-Watson matrix is
(
0 χ1(p) − 1χ2(p) − 1 0
)
. The
population dies out iff the spectral radius is 1. This gives the firstpart of
Theorem (Kozakova)
Let G = G1 ∗ G2. Then
1. For the free product with standard generating set, pc is the
smallest positive solution of (χ1(p) − 1)(χ2(p) − 1) = 1.
Kozakova’s results I
Let G = G1 ∗ G2. Then the Cayley graph is the free product ofgraphs of G1 and G2.Consider the branching process with two types of children. The
Galton-Watson matrix is
(
0 χ1(p) − 1χ2(p) − 1 0
)
. The
population dies out iff the spectral radius is 1. This gives the firstpart of
Theorem (Kozakova)
Let G = G1 ∗ G2. Then
1. For the free product with standard generating set, pc is the
smallest positive solution of (χ1(p) − 1)(χ2(p) − 1) = 1.
2. pexp(G1 ∗ G2) = minpexp(G1), pexp(G2).
The biggest known pc
In particular pc of PSL2(Z) is .51..., the biggest known pc of aCayley graph.
The biggest known pc
In particular pc of PSL2(Z) is .51..., the biggest known pc of aCayley graph.
Theorem (Kozakova-S)
There exists a quotient of PSL2(Z) with a strictly bigger pc .
The biggest known pc
In particular pc of PSL2(Z) is .51..., the biggest known pc of aCayley graph.
Theorem (Kozakova-S)
There exists a quotient of PSL2(Z) with a strictly bigger pc .
ProblemIs there a Cayley graph with pc > .6?
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
For a percolation with parameter p there exist a branchingprocess on the tree of pieces such that the expectedpopulation size is finite if and only if the expected cluster sizeof the percolation is finite.
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
For a percolation with parameter p there exist a branchingprocess on the tree of pieces such that the expectedpopulation size is finite if and only if the expected cluster sizeof the percolation is finite.
If all the border sets are finite then the branching process hasfinitely many types,
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
For a percolation with parameter p there exist a branchingprocess on the tree of pieces such that the expectedpopulation size is finite if and only if the expected cluster sizeof the percolation is finite.
If all the border sets are finite then the branching process hasfinitely many types, and the first moment matrix is of finitesize.
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
For a percolation with parameter p there exist a branchingprocess on the tree of pieces such that the expectedpopulation size is finite if and only if the expected cluster sizeof the percolation is finite.
If all the border sets are finite then the branching process hasfinitely many types, and the first moment matrix is of finitesize.
If, in addition, the pieces are finite,
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
For a percolation with parameter p there exist a branchingprocess on the tree of pieces such that the expectedpopulation size is finite if and only if the expected cluster sizeof the percolation is finite.
If all the border sets are finite then the branching process hasfinitely many types, and the first moment matrix is of finitesize.
If, in addition, the pieces are finite, then the entries of thefirst moment matrix are algebraic functions in p.
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
For a percolation with parameter p there exist a branchingprocess on the tree of pieces such that the expectedpopulation size is finite if and only if the expected cluster sizeof the percolation is finite.
If all the border sets are finite then the branching process hasfinitely many types, and the first moment matrix is of finitesize.
If, in addition, the pieces are finite, then the entries of thefirst moment matrix are algebraic functions in p. In this casepc is an algebraic number, which is the smallest value of p
such that the spectral radius of the first moment matrix is 1.
Kozakova’s results II
Assume a transitive graph X has a tree-like structure. Then
For a percolation with parameter p there exist a branchingprocess on the tree of pieces such that the expectedpopulation size is finite if and only if the expected cluster sizeof the percolation is finite.
If all the border sets are finite then the branching process hasfinitely many types, and the first moment matrix is of finitesize.
If, in addition, the pieces are finite, then the entries of thefirst moment matrix are algebraic functions in p. In this casepc is an algebraic number, which is the smallest value of p
such that the spectral radius of the first moment matrix is 1.Moreover there exists an algorithm, that, given the pieces ofthe tree-like structure, produces a finite extension K of thefield Q(x), and an algebraic function f (x) such that pc is thesmallest positive root of f (x).
Kozakova’s results III
TheoremThere exists an algorithm that, given a finite generating set of a
virtually free group, finds the pc of the corresponding Cayley graph.
Kozakova’s results III
TheoremThere exists an algorithm that, given a finite generating set of a
virtually free group, finds the pc of the corresponding Cayley graph.
This is the first quasi-isometry complete class of Cayley graphs ofgroups such that there exists an algorithm to find the pc of anygraph in the class (except the class of virtually cyclic groups wherepc is always 1).